In the Gibson Assembly reaction, there is an exonuclease reaction step in which the 5' ends of the dsDNA fragments are chewed back generating overlapping ends. Once the DNA fragments are incubated with the Gibson Assembly master mix, the overlapping ends will be generated that will subsequently be ligated.
For designing the overlapping ends, you can use the Picky thermodynamic analysis software as mentioned in this research article - http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0145682.
As few have mentioned before, you could use regular symmetric PCR to produce your fragments. Make sure to include extra base pairs on your primers to form the overlapping sequence to ligate to the adjacent "casettes" or vector sequence. The length of this overlapping region may affect the success of your assembly. Some have suggested 20, 40 to even of 100 bp.
You need to be ready with those DNA casettes to be inserted before proceeding with Gibson assembly.
In Gibson assembly, you need exonuclease that would chew away some part of your DNA. This enzyme starts "chewing" from 5' end and we can't control exactly how much of the DNA strand the enzyme would chew, so we need to somehow have a sense of timing when doing this assembly reaction. Without this "chewing" step, your DNA fragments (or casettes) that you produced separately from your PCR tubes in previous step would not combine with each other because the overlapping part has its complimentary strand from the regular PCR.
Don't worry if the exonuclease "chew" some part of your gene casettes, because DNA polymerase would add the complementary bases and ligase would "glue" all fragments in the assembly.